I publish a couple of open-source applications on VIA PicoITX platforms.
Though not strictly embedded, the machines are dedicated to the applications concerned.

The distribution is Slackware, and the distro, kernel, libraries, and apps have all been configured specifically for the target application and platform.

One of the applications is a non-real-time scheduler that depends on the fork(), execlp(), and waitpid() mechanism to sequentially manage automated application launch.

That mechanism fails when the machines use kernel versions 2.6.24 and later. The processes concerned do exit as expected, but the waitpid() function doesn't return.

I reverted to stock kernels and unmodified distrubutions (Slackware 12.0 and 12.1) to help rule out my customizations -- the problem persists unless I use a 2.6.23.x kernel.

The customized version performs well and as expected on Intel P4 platforms with the newer kernels.

So far, my debugging efforts have been cursory and limited to distro and kernel configuration and observations... that leads me to believe that something in the newer kernels peculiar to the C7 and/or VIA front end chips isn't playing nice. I still haven't ruled out libc or other potential culprits, however.

btw, the waitpid problem isn't the only one (I don't want to introduce a red herring, I don't know if this is related) -- USB mass storage also is fragile in the newer configurations. If the storage device has a journaled file system, the journal update is very slow and fails with alarming regularity. This problem tracks with the observations above, performing well on 2.6.23.x or with newer kernels and Intel hardware, and failing with newer kernels regardless of configuration on VIA platforms.

Hence this post... has anyone else here noticed unusual behavior on VIA hardware with newer kernels?